Torchsummary

Common errors

• RuntimeError: Given groups=1, weight of size [64, 3, 7, 7], expected input[1, 1, 224, 224] to have 3 channels, but got 1 channels instead

  cause The `input_size` passed to `summary()` does not match the number of input channels expected by the model's first layer.
  fix Adjust `input_size` to correctly reflect the `(channels, height, width)` of your input data. For example, if the model expects 3 channels, `input_size=(3, H, W)`.

• TypeError: summary() got an unexpected keyword argument 'input_data'

  cause This error typically occurs when attempting to use Keras-like argument names or non-existent arguments with `torchsummary.summary()`. The correct argument for specifying the input tensor shape (without the batch dimension) is `input_size`.
  fix Replace `input_data` or `input_shape` with `input_size`.

• AttributeError: 'torch.device' object has no attribute 'lower' or similar device-related error when passing device to summary()

  cause The `device` argument in `summary()` expects a string representation of the device (e.g., "cpu", "cuda"), not a `torch.device` object.
  fix Convert your `torch.device` object to a string before passing it: `device=str(your_torch_device_object)`.

Warnings

• gotcha The `input_size` parameter must be a tuple representing the shape of a *single input sample* (excluding the batch dimension). For example, a batch of 64 images of size (3, 224, 224) would require `input_size=(3, 224, 224)`. Mismatches are a common cause of `RuntimeError` or incorrect summaries.
  Fix: Carefully determine the expected input shape for your model's first layer, excluding the batch dimension, and pass it as a tuple to `input_size`.
• gotcha `torchsummary` internally uses `torch.jit.trace` to analyze the model. Models with dynamic control flow (e.g., if-statements or loops whose behavior depends on input data values) may not be correctly summarized or might raise `RuntimeError` during tracing.
  Fix: For models with highly dynamic control flow, consider using alternative tools like `torchinfo` (which uses a different tracing mechanism) or manually inspecting layer outputs.
• gotcha The `device` parameter in `summary()` expects a string (`"cpu"` or `"cuda"`) not a `torch.device` object directly.
  Fix: Convert your `torch.device` object to a string before passing it: `device=str(your_torch_device_object)`.

Install

• pip install torchsummary Install stable version

Imports

• summary
  wrong:

  from torchsummary.torchsummary import summary

  correct:

  from torchsummary import summary

  The 'summary' function is directly exposed at the top-level package.

Quickstart

This example defines a simple Convolutional Neural Network and uses `torchsummary.summary` to print its architecture, output shapes, and parameter counts. Note the importance of providing a correct `input_size` tuple (excluding batch dimension) and ensuring the model is on the specified device.

import torch
import torch.nn as nn
from torchsummary import summary

# Define a simple model
class SimpleCNN(nn.Module):
    def __init__(self):
        super(SimpleCNN, self).__init__()
        self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
        self.relu1 = nn.ReLU()
        self.pool1 = nn.MaxPool2d(kernel_size=2)
        self.flatten = nn.Flatten()
        # Calculate input size for linear layer: (28 - 5 + 1) / 2 = 12
        # So, 10 channels * 12 * 12 pixels = 1440
        self.fc1 = nn.Linear(10 * 12 * 12, 50)
        self.relu2 = nn.ReLU()
        self.fc2 = nn.Linear(50, 10)

    def forward(self, x):
        x = self.pool1(self.relu1(self.conv1(x)))
        x = self.flatten(x)
        x = self.relu2(self.fc1(x))
        x = self.fc2(x)
        return x

model = SimpleCNN()

# Determine device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

# Print summary for an input image of 1 channel, 28x28 pixels
# input_size should be (channels, height, width) *without* the batch dimension
print(summary(model, input_size=(1, 28, 28), device=str(device)))